xref: /openbmc/qemu/dump/dump.c (revision 125062e7)
1 /*
2  * QEMU dump
3  *
4  * Copyright Fujitsu, Corp. 2011, 2012
5  *
6  * Authors:
7  *     Wen Congyang <wency@cn.fujitsu.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
10  * See the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "qemu/bswap.h"
18 #include "exec/target_page.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/runstate.h"
22 #include "sysemu/cpus.h"
23 #include "qapi/error.h"
24 #include "qapi/qapi-commands-dump.h"
25 #include "qapi/qapi-events-dump.h"
26 #include "qapi/qmp/qerror.h"
27 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "hw/misc/vmcoreinfo.h"
30 #include "migration/blocker.h"
31 #include "hw/core/cpu.h"
32 #include "win_dump.h"
33 
34 #include <zlib.h>
35 #ifdef CONFIG_LZO
36 #include <lzo/lzo1x.h>
37 #endif
38 #ifdef CONFIG_SNAPPY
39 #include <snappy-c.h>
40 #endif
41 #ifndef ELF_MACHINE_UNAME
42 #define ELF_MACHINE_UNAME "Unknown"
43 #endif
44 
45 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
46 
47 static Error *dump_migration_blocker;
48 
49 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
50     ((DIV_ROUND_UP((hdr_size), 4) +                     \
51       DIV_ROUND_UP((name_size), 4) +                    \
52       DIV_ROUND_UP((desc_size), 4)) * 4)
53 
54 static inline bool dump_is_64bit(DumpState *s)
55 {
56     return s->dump_info.d_class == ELFCLASS64;
57 }
58 
59 static inline bool dump_has_filter(DumpState *s)
60 {
61     return s->filter_area_length > 0;
62 }
63 
64 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
65 {
66     if (s->dump_info.d_endian == ELFDATA2LSB) {
67         val = cpu_to_le16(val);
68     } else {
69         val = cpu_to_be16(val);
70     }
71 
72     return val;
73 }
74 
75 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
76 {
77     if (s->dump_info.d_endian == ELFDATA2LSB) {
78         val = cpu_to_le32(val);
79     } else {
80         val = cpu_to_be32(val);
81     }
82 
83     return val;
84 }
85 
86 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
87 {
88     if (s->dump_info.d_endian == ELFDATA2LSB) {
89         val = cpu_to_le64(val);
90     } else {
91         val = cpu_to_be64(val);
92     }
93 
94     return val;
95 }
96 
97 static int dump_cleanup(DumpState *s)
98 {
99     guest_phys_blocks_free(&s->guest_phys_blocks);
100     memory_mapping_list_free(&s->list);
101     close(s->fd);
102     g_free(s->guest_note);
103     g_clear_pointer(&s->string_table_buf, g_array_unref);
104     s->guest_note = NULL;
105     if (s->resume) {
106         if (s->detached) {
107             qemu_mutex_lock_iothread();
108         }
109         vm_start();
110         if (s->detached) {
111             qemu_mutex_unlock_iothread();
112         }
113     }
114     migrate_del_blocker(&dump_migration_blocker);
115 
116     return 0;
117 }
118 
119 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
120 {
121     DumpState *s = opaque;
122     size_t written_size;
123 
124     written_size = qemu_write_full(s->fd, buf, size);
125     if (written_size != size) {
126         return -errno;
127     }
128 
129     return 0;
130 }
131 
132 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
133 {
134     /*
135      * phnum in the elf header is 16 bit, if we have more segments we
136      * set phnum to PN_XNUM and write the real number of segments to a
137      * special section.
138      */
139     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
140 
141     memset(elf_header, 0, sizeof(Elf64_Ehdr));
142     memcpy(elf_header, ELFMAG, SELFMAG);
143     elf_header->e_ident[EI_CLASS] = ELFCLASS64;
144     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
145     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
146     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
147     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
148     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
149     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
150     elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
151     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
152     elf_header->e_phnum = cpu_to_dump16(s, phnum);
153     elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
154     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
155     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
156     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
157 }
158 
159 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
160 {
161     /*
162      * phnum in the elf header is 16 bit, if we have more segments we
163      * set phnum to PN_XNUM and write the real number of segments to a
164      * special section.
165      */
166     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
167 
168     memset(elf_header, 0, sizeof(Elf32_Ehdr));
169     memcpy(elf_header, ELFMAG, SELFMAG);
170     elf_header->e_ident[EI_CLASS] = ELFCLASS32;
171     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
172     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
173     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
174     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
175     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
176     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
177     elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
178     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
179     elf_header->e_phnum = cpu_to_dump16(s, phnum);
180     elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
181     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
182     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
183     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
184 }
185 
186 static void write_elf_header(DumpState *s, Error **errp)
187 {
188     Elf32_Ehdr elf32_header;
189     Elf64_Ehdr elf64_header;
190     size_t header_size;
191     void *header_ptr;
192     int ret;
193 
194     /* The NULL header and the shstrtab are always defined */
195     assert(s->shdr_num >= 2);
196     if (dump_is_64bit(s)) {
197         prepare_elf64_header(s, &elf64_header);
198         header_size = sizeof(elf64_header);
199         header_ptr = &elf64_header;
200     } else {
201         prepare_elf32_header(s, &elf32_header);
202         header_size = sizeof(elf32_header);
203         header_ptr = &elf32_header;
204     }
205 
206     ret = fd_write_vmcore(header_ptr, header_size, s);
207     if (ret < 0) {
208         error_setg_errno(errp, -ret, "dump: failed to write elf header");
209     }
210 }
211 
212 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
213                              int phdr_index, hwaddr offset,
214                              hwaddr filesz, Error **errp)
215 {
216     Elf64_Phdr phdr;
217     int ret;
218 
219     memset(&phdr, 0, sizeof(Elf64_Phdr));
220     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
221     phdr.p_offset = cpu_to_dump64(s, offset);
222     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
223     phdr.p_filesz = cpu_to_dump64(s, filesz);
224     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
225     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
226 
227     assert(memory_mapping->length >= filesz);
228 
229     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
230     if (ret < 0) {
231         error_setg_errno(errp, -ret,
232                          "dump: failed to write program header table");
233     }
234 }
235 
236 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
237                              int phdr_index, hwaddr offset,
238                              hwaddr filesz, Error **errp)
239 {
240     Elf32_Phdr phdr;
241     int ret;
242 
243     memset(&phdr, 0, sizeof(Elf32_Phdr));
244     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
245     phdr.p_offset = cpu_to_dump32(s, offset);
246     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
247     phdr.p_filesz = cpu_to_dump32(s, filesz);
248     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
249     phdr.p_vaddr =
250         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
251 
252     assert(memory_mapping->length >= filesz);
253 
254     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
255     if (ret < 0) {
256         error_setg_errno(errp, -ret,
257                          "dump: failed to write program header table");
258     }
259 }
260 
261 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
262 {
263     memset(phdr, 0, sizeof(*phdr));
264     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
265     phdr->p_offset = cpu_to_dump64(s, s->note_offset);
266     phdr->p_paddr = 0;
267     phdr->p_filesz = cpu_to_dump64(s, s->note_size);
268     phdr->p_memsz = cpu_to_dump64(s, s->note_size);
269     phdr->p_vaddr = 0;
270 }
271 
272 static inline int cpu_index(CPUState *cpu)
273 {
274     return cpu->cpu_index + 1;
275 }
276 
277 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
278                              Error **errp)
279 {
280     int ret;
281 
282     if (s->guest_note) {
283         ret = f(s->guest_note, s->guest_note_size, s);
284         if (ret < 0) {
285             error_setg(errp, "dump: failed to write guest note");
286         }
287     }
288 }
289 
290 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
291                               Error **errp)
292 {
293     CPUState *cpu;
294     int ret;
295     int id;
296 
297     CPU_FOREACH(cpu) {
298         id = cpu_index(cpu);
299         ret = cpu_write_elf64_note(f, cpu, id, s);
300         if (ret < 0) {
301             error_setg(errp, "dump: failed to write elf notes");
302             return;
303         }
304     }
305 
306     CPU_FOREACH(cpu) {
307         ret = cpu_write_elf64_qemunote(f, cpu, s);
308         if (ret < 0) {
309             error_setg(errp, "dump: failed to write CPU status");
310             return;
311         }
312     }
313 
314     write_guest_note(f, s, errp);
315 }
316 
317 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
318 {
319     memset(phdr, 0, sizeof(*phdr));
320     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
321     phdr->p_offset = cpu_to_dump32(s, s->note_offset);
322     phdr->p_paddr = 0;
323     phdr->p_filesz = cpu_to_dump32(s, s->note_size);
324     phdr->p_memsz = cpu_to_dump32(s, s->note_size);
325     phdr->p_vaddr = 0;
326 }
327 
328 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
329                               Error **errp)
330 {
331     CPUState *cpu;
332     int ret;
333     int id;
334 
335     CPU_FOREACH(cpu) {
336         id = cpu_index(cpu);
337         ret = cpu_write_elf32_note(f, cpu, id, s);
338         if (ret < 0) {
339             error_setg(errp, "dump: failed to write elf notes");
340             return;
341         }
342     }
343 
344     CPU_FOREACH(cpu) {
345         ret = cpu_write_elf32_qemunote(f, cpu, s);
346         if (ret < 0) {
347             error_setg(errp, "dump: failed to write CPU status");
348             return;
349         }
350     }
351 
352     write_guest_note(f, s, errp);
353 }
354 
355 static void write_elf_phdr_note(DumpState *s, Error **errp)
356 {
357     Elf32_Phdr phdr32;
358     Elf64_Phdr phdr64;
359     void *phdr;
360     size_t size;
361     int ret;
362 
363     if (dump_is_64bit(s)) {
364         prepare_elf64_phdr_note(s, &phdr64);
365         size = sizeof(phdr64);
366         phdr = &phdr64;
367     } else {
368         prepare_elf32_phdr_note(s, &phdr32);
369         size = sizeof(phdr32);
370         phdr = &phdr32;
371     }
372 
373     ret = fd_write_vmcore(phdr, size, s);
374     if (ret < 0) {
375         error_setg_errno(errp, -ret,
376                          "dump: failed to write program header table");
377     }
378 }
379 
380 static void prepare_elf_section_hdr_zero(DumpState *s)
381 {
382     if (dump_is_64bit(s)) {
383         Elf64_Shdr *shdr64 = s->elf_section_hdrs;
384 
385         shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
386     } else {
387         Elf32_Shdr *shdr32 = s->elf_section_hdrs;
388 
389         shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
390     }
391 }
392 
393 static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
394 {
395     uint64_t index = s->string_table_buf->len;
396     const char strtab[] = ".shstrtab";
397     Elf32_Shdr shdr32 = {};
398     Elf64_Shdr shdr64 = {};
399     int shdr_size;
400     void *shdr;
401 
402     g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
403     if (dump_is_64bit(s)) {
404         shdr_size = sizeof(Elf64_Shdr);
405         shdr64.sh_type = SHT_STRTAB;
406         shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
407         shdr64.sh_name = index;
408         shdr64.sh_size = s->string_table_buf->len;
409         shdr = &shdr64;
410     } else {
411         shdr_size = sizeof(Elf32_Shdr);
412         shdr32.sh_type = SHT_STRTAB;
413         shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
414         shdr32.sh_name = index;
415         shdr32.sh_size = s->string_table_buf->len;
416         shdr = &shdr32;
417     }
418     memcpy(buff, shdr, shdr_size);
419 }
420 
421 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
422 {
423     size_t len, sizeof_shdr;
424     void *buff_hdr;
425 
426     /*
427      * Section ordering:
428      * - HDR zero
429      * - Arch section hdrs
430      * - String table hdr
431      */
432     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
433     len = sizeof_shdr * s->shdr_num;
434     s->elf_section_hdrs = g_malloc0(len);
435     buff_hdr = s->elf_section_hdrs;
436 
437     /*
438      * The first section header is ALWAYS a special initial section
439      * header.
440      *
441      * The header should be 0 with one exception being that if
442      * phdr_num is PN_XNUM then the sh_info field contains the real
443      * number of segment entries.
444      *
445      * As we zero allocate the buffer we will only need to modify
446      * sh_info for the PN_XNUM case.
447      */
448     if (s->phdr_num >= PN_XNUM) {
449         prepare_elf_section_hdr_zero(s);
450     }
451     buff_hdr += sizeof_shdr;
452 
453     /* Add architecture defined section headers */
454     if (s->dump_info.arch_sections_write_hdr_fn
455         && s->shdr_num > 2) {
456         buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
457 
458         if (s->shdr_num >= SHN_LORESERVE) {
459             error_setg_errno(errp, EINVAL,
460                              "dump: too many architecture defined sections");
461             return false;
462         }
463     }
464 
465     /*
466      * String table is the last section since strings are added via
467      * arch_sections_write_hdr().
468      */
469     prepare_elf_section_hdr_string(s, buff_hdr);
470     return true;
471 }
472 
473 static void write_elf_section_headers(DumpState *s, Error **errp)
474 {
475     size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
476     int ret;
477 
478     if (!prepare_elf_section_hdrs(s, errp)) {
479         return;
480     }
481 
482     ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
483     if (ret < 0) {
484         error_setg_errno(errp, -ret, "dump: failed to write section headers");
485     }
486 
487     g_free(s->elf_section_hdrs);
488 }
489 
490 static void write_elf_sections(DumpState *s, Error **errp)
491 {
492     int ret;
493 
494     if (s->elf_section_data_size) {
495         /* Write architecture section data */
496         ret = fd_write_vmcore(s->elf_section_data,
497                               s->elf_section_data_size, s);
498         if (ret < 0) {
499             error_setg_errno(errp, -ret,
500                              "dump: failed to write architecture section data");
501             return;
502         }
503     }
504 
505     /* Write string table */
506     ret = fd_write_vmcore(s->string_table_buf->data,
507                           s->string_table_buf->len, s);
508     if (ret < 0) {
509         error_setg_errno(errp, -ret, "dump: failed to write string table data");
510     }
511 }
512 
513 static void write_data(DumpState *s, void *buf, int length, Error **errp)
514 {
515     int ret;
516 
517     ret = fd_write_vmcore(buf, length, s);
518     if (ret < 0) {
519         error_setg_errno(errp, -ret, "dump: failed to save memory");
520     } else {
521         s->written_size += length;
522     }
523 }
524 
525 /* write the memory to vmcore. 1 page per I/O. */
526 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
527                          int64_t size, Error **errp)
528 {
529     ERRP_GUARD();
530     int64_t i;
531 
532     for (i = 0; i < size / s->dump_info.page_size; i++) {
533         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
534                    s->dump_info.page_size, errp);
535         if (*errp) {
536             return;
537         }
538     }
539 
540     if ((size % s->dump_info.page_size) != 0) {
541         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
542                    size % s->dump_info.page_size, errp);
543         if (*errp) {
544             return;
545         }
546     }
547 }
548 
549 /* get the memory's offset and size in the vmcore */
550 static void get_offset_range(hwaddr phys_addr,
551                              ram_addr_t mapping_length,
552                              DumpState *s,
553                              hwaddr *p_offset,
554                              hwaddr *p_filesz)
555 {
556     GuestPhysBlock *block;
557     hwaddr offset = s->memory_offset;
558     int64_t size_in_block, start;
559 
560     /* When the memory is not stored into vmcore, offset will be -1 */
561     *p_offset = -1;
562     *p_filesz = 0;
563 
564     if (dump_has_filter(s)) {
565         if (phys_addr < s->filter_area_begin ||
566             phys_addr >= s->filter_area_begin + s->filter_area_length) {
567             return;
568         }
569     }
570 
571     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
572         if (dump_has_filter(s)) {
573             if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
574                 block->target_end <= s->filter_area_begin) {
575                 /* This block is out of the range */
576                 continue;
577             }
578 
579             if (s->filter_area_begin <= block->target_start) {
580                 start = block->target_start;
581             } else {
582                 start = s->filter_area_begin;
583             }
584 
585             size_in_block = block->target_end - start;
586             if (s->filter_area_begin + s->filter_area_length < block->target_end) {
587                 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
588             }
589         } else {
590             start = block->target_start;
591             size_in_block = block->target_end - block->target_start;
592         }
593 
594         if (phys_addr >= start && phys_addr < start + size_in_block) {
595             *p_offset = phys_addr - start + offset;
596 
597             /* The offset range mapped from the vmcore file must not spill over
598              * the GuestPhysBlock, clamp it. The rest of the mapping will be
599              * zero-filled in memory at load time; see
600              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
601              */
602             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
603                         mapping_length :
604                         size_in_block - (phys_addr - start);
605             return;
606         }
607 
608         offset += size_in_block;
609     }
610 }
611 
612 static void write_elf_phdr_loads(DumpState *s, Error **errp)
613 {
614     ERRP_GUARD();
615     hwaddr offset, filesz;
616     MemoryMapping *memory_mapping;
617     uint32_t phdr_index = 1;
618 
619     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
620         get_offset_range(memory_mapping->phys_addr,
621                          memory_mapping->length,
622                          s, &offset, &filesz);
623         if (dump_is_64bit(s)) {
624             write_elf64_load(s, memory_mapping, phdr_index++, offset,
625                              filesz, errp);
626         } else {
627             write_elf32_load(s, memory_mapping, phdr_index++, offset,
628                              filesz, errp);
629         }
630 
631         if (*errp) {
632             return;
633         }
634 
635         if (phdr_index >= s->phdr_num) {
636             break;
637         }
638     }
639 }
640 
641 static void write_elf_notes(DumpState *s, Error **errp)
642 {
643     if (dump_is_64bit(s)) {
644         write_elf64_notes(fd_write_vmcore, s, errp);
645     } else {
646         write_elf32_notes(fd_write_vmcore, s, errp);
647     }
648 }
649 
650 /* write elf header, PT_NOTE and elf note to vmcore. */
651 static void dump_begin(DumpState *s, Error **errp)
652 {
653     ERRP_GUARD();
654 
655     /*
656      * the vmcore's format is:
657      *   --------------
658      *   |  elf header |
659      *   --------------
660      *   |  sctn_hdr   |
661      *   --------------
662      *   |  PT_NOTE    |
663      *   --------------
664      *   |  PT_LOAD    |
665      *   --------------
666      *   |  ......     |
667      *   --------------
668      *   |  PT_LOAD    |
669      *   --------------
670      *   |  elf note   |
671      *   --------------
672      *   |  memory     |
673      *   --------------
674      *
675      * we only know where the memory is saved after we write elf note into
676      * vmcore.
677      */
678 
679     /* write elf header to vmcore */
680     write_elf_header(s, errp);
681     if (*errp) {
682         return;
683     }
684 
685     /* write section headers to vmcore */
686     write_elf_section_headers(s, errp);
687     if (*errp) {
688         return;
689     }
690 
691     /* write PT_NOTE to vmcore */
692     write_elf_phdr_note(s, errp);
693     if (*errp) {
694         return;
695     }
696 
697     /* write all PT_LOADs to vmcore */
698     write_elf_phdr_loads(s, errp);
699     if (*errp) {
700         return;
701     }
702 
703     /* write notes to vmcore */
704     write_elf_notes(s, errp);
705 }
706 
707 int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
708                                     int64_t filter_area_start,
709                                     int64_t filter_area_length)
710 {
711     int64_t size, left, right;
712 
713     /* No filter, return full size */
714     if (!filter_area_length) {
715         return block->target_end - block->target_start;
716     }
717 
718     /* calculate the overlapped region. */
719     left = MAX(filter_area_start, block->target_start);
720     right = MIN(filter_area_start + filter_area_length, block->target_end);
721     size = right - left;
722     size = size > 0 ? size : 0;
723 
724     return size;
725 }
726 
727 int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
728                                      int64_t filter_area_start,
729                                      int64_t filter_area_length)
730 {
731     if (filter_area_length) {
732         /* return -1 if the block is not within filter area */
733         if (block->target_start >= filter_area_start + filter_area_length ||
734             block->target_end <= filter_area_start) {
735             return -1;
736         }
737 
738         if (filter_area_start > block->target_start) {
739             return filter_area_start - block->target_start;
740         }
741     }
742 
743     return 0;
744 }
745 
746 /* write all memory to vmcore */
747 static void dump_iterate(DumpState *s, Error **errp)
748 {
749     ERRP_GUARD();
750     GuestPhysBlock *block;
751     int64_t memblock_size, memblock_start;
752 
753     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
754         memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
755         if (memblock_start == -1) {
756             continue;
757         }
758 
759         memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
760 
761         /* Write the memory to file */
762         write_memory(s, block, memblock_start, memblock_size, errp);
763         if (*errp) {
764             return;
765         }
766     }
767 }
768 
769 static void dump_end(DumpState *s, Error **errp)
770 {
771     int rc;
772 
773     if (s->elf_section_data_size) {
774         s->elf_section_data = g_malloc0(s->elf_section_data_size);
775     }
776 
777     /* Adds the architecture defined section data to s->elf_section_data  */
778     if (s->dump_info.arch_sections_write_fn &&
779         s->elf_section_data_size) {
780         rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
781         if (rc) {
782             error_setg_errno(errp, rc,
783                              "dump: failed to get arch section data");
784             g_free(s->elf_section_data);
785             return;
786         }
787     }
788 
789     /* write sections to vmcore */
790     write_elf_sections(s, errp);
791 }
792 
793 static void create_vmcore(DumpState *s, Error **errp)
794 {
795     ERRP_GUARD();
796 
797     dump_begin(s, errp);
798     if (*errp) {
799         return;
800     }
801 
802     /* Iterate over memory and dump it to file */
803     dump_iterate(s, errp);
804     if (*errp) {
805         return;
806     }
807 
808     /* Write the section data */
809     dump_end(s, errp);
810 }
811 
812 static int write_start_flat_header(DumpState *s)
813 {
814     MakedumpfileHeader *mh;
815     int ret = 0;
816 
817     if (s->kdump_raw) {
818         return 0;
819     }
820 
821     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
822     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
823 
824     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
825            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
826 
827     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
828     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
829 
830     size_t written_size;
831     written_size = qemu_write_full(s->fd, mh, MAX_SIZE_MDF_HEADER);
832     if (written_size != MAX_SIZE_MDF_HEADER) {
833         ret = -1;
834     }
835 
836     g_free(mh);
837     return ret;
838 }
839 
840 static int write_end_flat_header(DumpState *s)
841 {
842     MakedumpfileDataHeader mdh;
843 
844     if (s->kdump_raw) {
845         return 0;
846     }
847 
848     mdh.offset = END_FLAG_FLAT_HEADER;
849     mdh.buf_size = END_FLAG_FLAT_HEADER;
850 
851     size_t written_size;
852     written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
853     if (written_size != sizeof(mdh)) {
854         return -1;
855     }
856 
857     return 0;
858 }
859 
860 static int write_buffer(DumpState *s, off_t offset, const void *buf, size_t size)
861 {
862     size_t written_size;
863     MakedumpfileDataHeader mdh;
864     off_t seek_loc;
865 
866     if (s->kdump_raw) {
867         seek_loc = lseek(s->fd, offset, SEEK_SET);
868         if (seek_loc == (off_t) -1) {
869             return -1;
870         }
871     } else {
872         mdh.offset = cpu_to_be64(offset);
873         mdh.buf_size = cpu_to_be64(size);
874 
875         written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
876         if (written_size != sizeof(mdh)) {
877             return -1;
878         }
879     }
880 
881     written_size = qemu_write_full(s->fd, buf, size);
882     if (written_size != size) {
883         return -1;
884     }
885 
886     return 0;
887 }
888 
889 static int buf_write_note(const void *buf, size_t size, void *opaque)
890 {
891     DumpState *s = opaque;
892 
893     /* note_buf is not enough */
894     if (s->note_buf_offset + size > s->note_size) {
895         return -1;
896     }
897 
898     memcpy(s->note_buf + s->note_buf_offset, buf, size);
899 
900     s->note_buf_offset += size;
901 
902     return 0;
903 }
904 
905 /*
906  * This function retrieves various sizes from an elf header.
907  *
908  * @note has to be a valid ELF note. The return sizes are unmodified
909  * (not padded or rounded up to be multiple of 4).
910  */
911 static void get_note_sizes(DumpState *s, const void *note,
912                            uint64_t *note_head_size,
913                            uint64_t *name_size,
914                            uint64_t *desc_size)
915 {
916     uint64_t note_head_sz;
917     uint64_t name_sz;
918     uint64_t desc_sz;
919 
920     if (dump_is_64bit(s)) {
921         const Elf64_Nhdr *hdr = note;
922         note_head_sz = sizeof(Elf64_Nhdr);
923         name_sz = cpu_to_dump64(s, hdr->n_namesz);
924         desc_sz = cpu_to_dump64(s, hdr->n_descsz);
925     } else {
926         const Elf32_Nhdr *hdr = note;
927         note_head_sz = sizeof(Elf32_Nhdr);
928         name_sz = cpu_to_dump32(s, hdr->n_namesz);
929         desc_sz = cpu_to_dump32(s, hdr->n_descsz);
930     }
931 
932     if (note_head_size) {
933         *note_head_size = note_head_sz;
934     }
935     if (name_size) {
936         *name_size = name_sz;
937     }
938     if (desc_size) {
939         *desc_size = desc_sz;
940     }
941 }
942 
943 static bool note_name_equal(DumpState *s,
944                             const uint8_t *note, const char *name)
945 {
946     int len = strlen(name) + 1;
947     uint64_t head_size, name_size;
948 
949     get_note_sizes(s, note, &head_size, &name_size, NULL);
950     head_size = ROUND_UP(head_size, 4);
951 
952     return name_size == len && memcmp(note + head_size, name, len) == 0;
953 }
954 
955 /* write common header, sub header and elf note to vmcore */
956 static void create_header32(DumpState *s, Error **errp)
957 {
958     ERRP_GUARD();
959     DiskDumpHeader32 *dh = NULL;
960     KdumpSubHeader32 *kh = NULL;
961     size_t size;
962     uint32_t block_size;
963     uint32_t sub_hdr_size;
964     uint32_t bitmap_blocks;
965     uint32_t status = 0;
966     uint64_t offset_note;
967 
968     /* write common header, the version of kdump-compressed format is 6th */
969     size = sizeof(DiskDumpHeader32);
970     dh = g_malloc0(size);
971 
972     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
973     dh->header_version = cpu_to_dump32(s, 6);
974     block_size = s->dump_info.page_size;
975     dh->block_size = cpu_to_dump32(s, block_size);
976     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
977     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
978     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
979     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
980     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
981     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
982     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
983     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
984     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
985 
986     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
987         status |= DUMP_DH_COMPRESSED_ZLIB;
988     }
989 #ifdef CONFIG_LZO
990     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
991         status |= DUMP_DH_COMPRESSED_LZO;
992     }
993 #endif
994 #ifdef CONFIG_SNAPPY
995     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
996         status |= DUMP_DH_COMPRESSED_SNAPPY;
997     }
998 #endif
999     dh->status = cpu_to_dump32(s, status);
1000 
1001     if (write_buffer(s, 0, dh, size) < 0) {
1002         error_setg(errp, "dump: failed to write disk dump header");
1003         goto out;
1004     }
1005 
1006     /* write sub header */
1007     size = sizeof(KdumpSubHeader32);
1008     kh = g_malloc0(size);
1009 
1010     /* 64bit max_mapnr_64 */
1011     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1012     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1013     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1014 
1015     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1016     if (s->guest_note &&
1017         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1018         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1019 
1020         get_note_sizes(s, s->guest_note,
1021                        &hsize, &name_size, &size_vmcoreinfo_desc);
1022         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1023             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1024         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1025         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1026     }
1027 
1028     kh->offset_note = cpu_to_dump64(s, offset_note);
1029     kh->note_size = cpu_to_dump32(s, s->note_size);
1030 
1031     if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1032                      block_size, kh, size) < 0) {
1033         error_setg(errp, "dump: failed to write kdump sub header");
1034         goto out;
1035     }
1036 
1037     /* write note */
1038     s->note_buf = g_malloc0(s->note_size);
1039     s->note_buf_offset = 0;
1040 
1041     /* use s->note_buf to store notes temporarily */
1042     write_elf32_notes(buf_write_note, s, errp);
1043     if (*errp) {
1044         goto out;
1045     }
1046     if (write_buffer(s, offset_note, s->note_buf,
1047                      s->note_size) < 0) {
1048         error_setg(errp, "dump: failed to write notes");
1049         goto out;
1050     }
1051 
1052     /* get offset of dump_bitmap */
1053     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1054                              block_size;
1055 
1056     /* get offset of page */
1057     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1058                      block_size;
1059 
1060 out:
1061     g_free(dh);
1062     g_free(kh);
1063     g_free(s->note_buf);
1064 }
1065 
1066 /* write common header, sub header and elf note to vmcore */
1067 static void create_header64(DumpState *s, Error **errp)
1068 {
1069     ERRP_GUARD();
1070     DiskDumpHeader64 *dh = NULL;
1071     KdumpSubHeader64 *kh = NULL;
1072     size_t size;
1073     uint32_t block_size;
1074     uint32_t sub_hdr_size;
1075     uint32_t bitmap_blocks;
1076     uint32_t status = 0;
1077     uint64_t offset_note;
1078 
1079     /* write common header, the version of kdump-compressed format is 6th */
1080     size = sizeof(DiskDumpHeader64);
1081     dh = g_malloc0(size);
1082 
1083     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1084     dh->header_version = cpu_to_dump32(s, 6);
1085     block_size = s->dump_info.page_size;
1086     dh->block_size = cpu_to_dump32(s, block_size);
1087     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1088     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1089     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1090     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1091     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1092     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1093     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1094     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1095     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1096 
1097     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1098         status |= DUMP_DH_COMPRESSED_ZLIB;
1099     }
1100 #ifdef CONFIG_LZO
1101     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1102         status |= DUMP_DH_COMPRESSED_LZO;
1103     }
1104 #endif
1105 #ifdef CONFIG_SNAPPY
1106     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1107         status |= DUMP_DH_COMPRESSED_SNAPPY;
1108     }
1109 #endif
1110     dh->status = cpu_to_dump32(s, status);
1111 
1112     if (write_buffer(s, 0, dh, size) < 0) {
1113         error_setg(errp, "dump: failed to write disk dump header");
1114         goto out;
1115     }
1116 
1117     /* write sub header */
1118     size = sizeof(KdumpSubHeader64);
1119     kh = g_malloc0(size);
1120 
1121     /* 64bit max_mapnr_64 */
1122     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1123     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1124     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1125 
1126     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1127     if (s->guest_note &&
1128         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1129         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1130 
1131         get_note_sizes(s, s->guest_note,
1132                        &hsize, &name_size, &size_vmcoreinfo_desc);
1133         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1134             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1135         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1136         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1137     }
1138 
1139     kh->offset_note = cpu_to_dump64(s, offset_note);
1140     kh->note_size = cpu_to_dump64(s, s->note_size);
1141 
1142     if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1143                      block_size, kh, size) < 0) {
1144         error_setg(errp, "dump: failed to write kdump sub header");
1145         goto out;
1146     }
1147 
1148     /* write note */
1149     s->note_buf = g_malloc0(s->note_size);
1150     s->note_buf_offset = 0;
1151 
1152     /* use s->note_buf to store notes temporarily */
1153     write_elf64_notes(buf_write_note, s, errp);
1154     if (*errp) {
1155         goto out;
1156     }
1157 
1158     if (write_buffer(s, offset_note, s->note_buf,
1159                      s->note_size) < 0) {
1160         error_setg(errp, "dump: failed to write notes");
1161         goto out;
1162     }
1163 
1164     /* get offset of dump_bitmap */
1165     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1166                              block_size;
1167 
1168     /* get offset of page */
1169     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1170                      block_size;
1171 
1172 out:
1173     g_free(dh);
1174     g_free(kh);
1175     g_free(s->note_buf);
1176 }
1177 
1178 static void write_dump_header(DumpState *s, Error **errp)
1179 {
1180     if (dump_is_64bit(s)) {
1181         create_header64(s, errp);
1182     } else {
1183         create_header32(s, errp);
1184     }
1185 }
1186 
1187 static size_t dump_bitmap_get_bufsize(DumpState *s)
1188 {
1189     return s->dump_info.page_size;
1190 }
1191 
1192 /*
1193  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1194  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1195  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1196  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1197  * vmcore, ie. synchronizing un-sync bit into vmcore.
1198  */
1199 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1200                            uint8_t *buf, DumpState *s)
1201 {
1202     off_t old_offset, new_offset;
1203     off_t offset_bitmap1, offset_bitmap2;
1204     uint32_t byte, bit;
1205     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1206     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1207 
1208     /* should not set the previous place */
1209     assert(last_pfn <= pfn);
1210 
1211     /*
1212      * if the bit needed to be set is not cached in buf, flush the data in buf
1213      * to vmcore firstly.
1214      * making new_offset be bigger than old_offset can also sync remained data
1215      * into vmcore.
1216      */
1217     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1218     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1219 
1220     while (old_offset < new_offset) {
1221         /* calculate the offset and write dump_bitmap */
1222         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1223         if (write_buffer(s, offset_bitmap1, buf,
1224                          bitmap_bufsize) < 0) {
1225             return -1;
1226         }
1227 
1228         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1229         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1230                          old_offset;
1231         if (write_buffer(s, offset_bitmap2, buf,
1232                          bitmap_bufsize) < 0) {
1233             return -1;
1234         }
1235 
1236         memset(buf, 0, bitmap_bufsize);
1237         old_offset += bitmap_bufsize;
1238     }
1239 
1240     /* get the exact place of the bit in the buf, and set it */
1241     byte = (pfn % bits_per_buf) / CHAR_BIT;
1242     bit = (pfn % bits_per_buf) % CHAR_BIT;
1243     if (value) {
1244         buf[byte] |= 1u << bit;
1245     } else {
1246         buf[byte] &= ~(1u << bit);
1247     }
1248 
1249     return 0;
1250 }
1251 
1252 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1253 {
1254     int target_page_shift = ctz32(s->dump_info.page_size);
1255 
1256     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1257 }
1258 
1259 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1260 {
1261     int target_page_shift = ctz32(s->dump_info.page_size);
1262 
1263     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1264 }
1265 
1266 /*
1267  * Return the page frame number and the page content in *bufptr. bufptr can be
1268  * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1269  * memory. This is not necessarily the memory returned.
1270  */
1271 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1272                           uint8_t **bufptr, DumpState *s)
1273 {
1274     GuestPhysBlock *block = *blockptr;
1275     uint32_t page_size = s->dump_info.page_size;
1276     uint8_t *buf = NULL, *hbuf;
1277     hwaddr addr;
1278 
1279     /* block == NULL means the start of the iteration */
1280     if (!block) {
1281         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1282         *blockptr = block;
1283         addr = block->target_start;
1284         *pfnptr = dump_paddr_to_pfn(s, addr);
1285     } else {
1286         *pfnptr += 1;
1287         addr = dump_pfn_to_paddr(s, *pfnptr);
1288     }
1289     assert(block != NULL);
1290 
1291     while (1) {
1292         if (addr >= block->target_start && addr < block->target_end) {
1293             size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1294             hbuf = block->host_addr + (addr - block->target_start);
1295             if (!buf) {
1296                 if (n == page_size) {
1297                     /* this is a whole target page, go for it */
1298                     assert(addr % page_size == 0);
1299                     buf = hbuf;
1300                     break;
1301                 } else if (bufptr) {
1302                     assert(*bufptr);
1303                     buf = *bufptr;
1304                     memset(buf, 0, page_size);
1305                 } else {
1306                     return true;
1307                 }
1308             }
1309 
1310             memcpy(buf + addr % page_size, hbuf, n);
1311             addr += n;
1312             if (addr % page_size == 0 || addr >= block->target_end) {
1313                 /* we filled up the page or the current block is finished */
1314                 break;
1315             }
1316         } else {
1317             /* the next page is in the next block */
1318             *blockptr = block = QTAILQ_NEXT(block, next);
1319             if (!block) {
1320                 break;
1321             }
1322 
1323             addr = block->target_start;
1324             /* are we still in the same page? */
1325             if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1326                 if (buf) {
1327                     /* no, but we already filled something earlier, return it */
1328                     break;
1329                 } else {
1330                     /* else continue from there */
1331                     *pfnptr = dump_paddr_to_pfn(s, addr);
1332                 }
1333             }
1334         }
1335     }
1336 
1337     if (bufptr) {
1338         *bufptr = buf;
1339     }
1340 
1341     return buf != NULL;
1342 }
1343 
1344 static void write_dump_bitmap(DumpState *s, Error **errp)
1345 {
1346     int ret = 0;
1347     uint64_t last_pfn, pfn;
1348     void *dump_bitmap_buf;
1349     size_t num_dumpable;
1350     GuestPhysBlock *block_iter = NULL;
1351     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1352     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1353 
1354     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1355     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1356 
1357     num_dumpable = 0;
1358     last_pfn = 0;
1359 
1360     /*
1361      * exam memory page by page, and set the bit in dump_bitmap corresponded
1362      * to the existing page.
1363      */
1364     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1365         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1366         if (ret < 0) {
1367             error_setg(errp, "dump: failed to set dump_bitmap");
1368             goto out;
1369         }
1370 
1371         last_pfn = pfn;
1372         num_dumpable++;
1373     }
1374 
1375     /*
1376      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1377      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1378      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1379      */
1380     if (num_dumpable > 0) {
1381         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1382                               dump_bitmap_buf, s);
1383         if (ret < 0) {
1384             error_setg(errp, "dump: failed to sync dump_bitmap");
1385             goto out;
1386         }
1387     }
1388 
1389     /* number of dumpable pages that will be dumped later */
1390     s->num_dumpable = num_dumpable;
1391 
1392 out:
1393     g_free(dump_bitmap_buf);
1394 }
1395 
1396 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1397                                off_t offset)
1398 {
1399     data_cache->state = s;
1400     data_cache->data_size = 0;
1401     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1402     data_cache->buf = g_malloc0(data_cache->buf_size);
1403     data_cache->offset = offset;
1404 }
1405 
1406 static int write_cache(DataCache *dc, const void *buf, size_t size,
1407                        bool flag_sync)
1408 {
1409     /*
1410      * dc->buf_size should not be less than size, otherwise dc will never be
1411      * enough
1412      */
1413     assert(size <= dc->buf_size);
1414 
1415     /*
1416      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1417      * otherwise check if the space is enough for caching data in buf, if not,
1418      * write the data in dc->buf to dc->state->fd and reset dc->buf
1419      */
1420     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1421         (flag_sync && dc->data_size > 0)) {
1422         if (write_buffer(dc->state, dc->offset, dc->buf, dc->data_size) < 0) {
1423             return -1;
1424         }
1425 
1426         dc->offset += dc->data_size;
1427         dc->data_size = 0;
1428     }
1429 
1430     if (!flag_sync) {
1431         memcpy(dc->buf + dc->data_size, buf, size);
1432         dc->data_size += size;
1433     }
1434 
1435     return 0;
1436 }
1437 
1438 static void free_data_cache(DataCache *data_cache)
1439 {
1440     g_free(data_cache->buf);
1441 }
1442 
1443 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1444 {
1445     switch (flag_compress) {
1446     case DUMP_DH_COMPRESSED_ZLIB:
1447         return compressBound(page_size);
1448 
1449     case DUMP_DH_COMPRESSED_LZO:
1450         /*
1451          * LZO will expand incompressible data by a little amount. Please check
1452          * the following URL to see the expansion calculation:
1453          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1454          */
1455         return page_size + page_size / 16 + 64 + 3;
1456 
1457 #ifdef CONFIG_SNAPPY
1458     case DUMP_DH_COMPRESSED_SNAPPY:
1459         return snappy_max_compressed_length(page_size);
1460 #endif
1461     }
1462     return 0;
1463 }
1464 
1465 static void write_dump_pages(DumpState *s, Error **errp)
1466 {
1467     int ret = 0;
1468     DataCache page_desc, page_data;
1469     size_t len_buf_out, size_out;
1470 #ifdef CONFIG_LZO
1471     lzo_bytep wrkmem = NULL;
1472 #endif
1473     uint8_t *buf_out = NULL;
1474     off_t offset_desc, offset_data;
1475     PageDescriptor pd, pd_zero;
1476     uint8_t *buf;
1477     GuestPhysBlock *block_iter = NULL;
1478     uint64_t pfn_iter;
1479     g_autofree uint8_t *page = NULL;
1480 
1481     /* get offset of page_desc and page_data in dump file */
1482     offset_desc = s->offset_page;
1483     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1484 
1485     prepare_data_cache(&page_desc, s, offset_desc);
1486     prepare_data_cache(&page_data, s, offset_data);
1487 
1488     /* prepare buffer to store compressed data */
1489     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1490     assert(len_buf_out != 0);
1491 
1492 #ifdef CONFIG_LZO
1493     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1494 #endif
1495 
1496     buf_out = g_malloc(len_buf_out);
1497 
1498     /*
1499      * init zero page's page_desc and page_data, because every zero page
1500      * uses the same page_data
1501      */
1502     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1503     pd_zero.flags = cpu_to_dump32(s, 0);
1504     pd_zero.offset = cpu_to_dump64(s, offset_data);
1505     pd_zero.page_flags = cpu_to_dump64(s, 0);
1506     buf = g_malloc0(s->dump_info.page_size);
1507     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1508     g_free(buf);
1509     if (ret < 0) {
1510         error_setg(errp, "dump: failed to write page data (zero page)");
1511         goto out;
1512     }
1513 
1514     offset_data += s->dump_info.page_size;
1515     page = g_malloc(s->dump_info.page_size);
1516 
1517     /*
1518      * dump memory to vmcore page by page. zero page will all be resided in the
1519      * first page of page section
1520      */
1521     for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1522         /* check zero page */
1523         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1524             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1525                               false);
1526             if (ret < 0) {
1527                 error_setg(errp, "dump: failed to write page desc");
1528                 goto out;
1529             }
1530         } else {
1531             /*
1532              * not zero page, then:
1533              * 1. compress the page
1534              * 2. write the compressed page into the cache of page_data
1535              * 3. get page desc of the compressed page and write it into the
1536              *    cache of page_desc
1537              *
1538              * only one compression format will be used here, for
1539              * s->flag_compress is set. But when compression fails to work,
1540              * we fall back to save in plaintext.
1541              */
1542              size_out = len_buf_out;
1543              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1544                     (compress2(buf_out, (uLongf *)&size_out, buf,
1545                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1546                     (size_out < s->dump_info.page_size)) {
1547                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1548                 pd.size  = cpu_to_dump32(s, size_out);
1549 
1550                 ret = write_cache(&page_data, buf_out, size_out, false);
1551                 if (ret < 0) {
1552                     error_setg(errp, "dump: failed to write page data");
1553                     goto out;
1554                 }
1555 #ifdef CONFIG_LZO
1556             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1557                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1558                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1559                     (size_out < s->dump_info.page_size)) {
1560                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1561                 pd.size  = cpu_to_dump32(s, size_out);
1562 
1563                 ret = write_cache(&page_data, buf_out, size_out, false);
1564                 if (ret < 0) {
1565                     error_setg(errp, "dump: failed to write page data");
1566                     goto out;
1567                 }
1568 #endif
1569 #ifdef CONFIG_SNAPPY
1570             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1571                     (snappy_compress((char *)buf, s->dump_info.page_size,
1572                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1573                     (size_out < s->dump_info.page_size)) {
1574                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1575                 pd.size  = cpu_to_dump32(s, size_out);
1576 
1577                 ret = write_cache(&page_data, buf_out, size_out, false);
1578                 if (ret < 0) {
1579                     error_setg(errp, "dump: failed to write page data");
1580                     goto out;
1581                 }
1582 #endif
1583             } else {
1584                 /*
1585                  * fall back to save in plaintext, size_out should be
1586                  * assigned the target's page size
1587                  */
1588                 pd.flags = cpu_to_dump32(s, 0);
1589                 size_out = s->dump_info.page_size;
1590                 pd.size = cpu_to_dump32(s, size_out);
1591 
1592                 ret = write_cache(&page_data, buf,
1593                                   s->dump_info.page_size, false);
1594                 if (ret < 0) {
1595                     error_setg(errp, "dump: failed to write page data");
1596                     goto out;
1597                 }
1598             }
1599 
1600             /* get and write page desc here */
1601             pd.page_flags = cpu_to_dump64(s, 0);
1602             pd.offset = cpu_to_dump64(s, offset_data);
1603             offset_data += size_out;
1604 
1605             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1606             if (ret < 0) {
1607                 error_setg(errp, "dump: failed to write page desc");
1608                 goto out;
1609             }
1610         }
1611         s->written_size += s->dump_info.page_size;
1612     }
1613 
1614     ret = write_cache(&page_desc, NULL, 0, true);
1615     if (ret < 0) {
1616         error_setg(errp, "dump: failed to sync cache for page_desc");
1617         goto out;
1618     }
1619     ret = write_cache(&page_data, NULL, 0, true);
1620     if (ret < 0) {
1621         error_setg(errp, "dump: failed to sync cache for page_data");
1622         goto out;
1623     }
1624 
1625 out:
1626     free_data_cache(&page_desc);
1627     free_data_cache(&page_data);
1628 
1629 #ifdef CONFIG_LZO
1630     g_free(wrkmem);
1631 #endif
1632 
1633     g_free(buf_out);
1634 }
1635 
1636 static void create_kdump_vmcore(DumpState *s, Error **errp)
1637 {
1638     ERRP_GUARD();
1639     int ret;
1640 
1641     /*
1642      * the kdump-compressed format is:
1643      *                                               File offset
1644      *  +------------------------------------------+ 0x0
1645      *  |    main header (struct disk_dump_header) |
1646      *  |------------------------------------------+ block 1
1647      *  |    sub header (struct kdump_sub_header)  |
1648      *  |------------------------------------------+ block 2
1649      *  |            1st-dump_bitmap               |
1650      *  |------------------------------------------+ block 2 + X blocks
1651      *  |            2nd-dump_bitmap               | (aligned by block)
1652      *  |------------------------------------------+ block 2 + 2 * X blocks
1653      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1654      *  |  page desc for pfn 1 (struct page_desc)  |
1655      *  |                    :                     |
1656      *  |------------------------------------------| (not aligned by block)
1657      *  |         page data (pfn 0)                |
1658      *  |         page data (pfn 1)                |
1659      *  |                    :                     |
1660      *  +------------------------------------------+
1661      */
1662 
1663     ret = write_start_flat_header(s);
1664     if (ret < 0) {
1665         error_setg(errp, "dump: failed to write start flat header");
1666         return;
1667     }
1668 
1669     write_dump_header(s, errp);
1670     if (*errp) {
1671         return;
1672     }
1673 
1674     write_dump_bitmap(s, errp);
1675     if (*errp) {
1676         return;
1677     }
1678 
1679     write_dump_pages(s, errp);
1680     if (*errp) {
1681         return;
1682     }
1683 
1684     ret = write_end_flat_header(s);
1685     if (ret < 0) {
1686         error_setg(errp, "dump: failed to write end flat header");
1687         return;
1688     }
1689 }
1690 
1691 static void get_max_mapnr(DumpState *s)
1692 {
1693     GuestPhysBlock *last_block;
1694 
1695     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1696     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1697 }
1698 
1699 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1700 
1701 static void dump_state_prepare(DumpState *s)
1702 {
1703     /* zero the struct, setting status to active */
1704     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1705 }
1706 
1707 bool qemu_system_dump_in_progress(void)
1708 {
1709     DumpState *state = &dump_state_global;
1710     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1711 }
1712 
1713 /*
1714  * calculate total size of memory to be dumped (taking filter into
1715  * account.)
1716  */
1717 static int64_t dump_calculate_size(DumpState *s)
1718 {
1719     GuestPhysBlock *block;
1720     int64_t total = 0;
1721 
1722     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1723         total += dump_filtered_memblock_size(block,
1724                                              s->filter_area_begin,
1725                                              s->filter_area_length);
1726     }
1727 
1728     return total;
1729 }
1730 
1731 static void vmcoreinfo_update_phys_base(DumpState *s)
1732 {
1733     uint64_t size, note_head_size, name_size, phys_base;
1734     char **lines;
1735     uint8_t *vmci;
1736     size_t i;
1737 
1738     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1739         return;
1740     }
1741 
1742     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1743     note_head_size = ROUND_UP(note_head_size, 4);
1744 
1745     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1746     *(vmci + size) = '\0';
1747 
1748     lines = g_strsplit((char *)vmci, "\n", -1);
1749     for (i = 0; lines[i]; i++) {
1750         const char *prefix = NULL;
1751 
1752         if (s->dump_info.d_machine == EM_X86_64) {
1753             prefix = "NUMBER(phys_base)=";
1754         } else if (s->dump_info.d_machine == EM_AARCH64) {
1755             prefix = "NUMBER(PHYS_OFFSET)=";
1756         }
1757 
1758         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1759             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1760                               &phys_base) < 0) {
1761                 warn_report("Failed to read %s", prefix);
1762             } else {
1763                 s->dump_info.phys_base = phys_base;
1764             }
1765             break;
1766         }
1767     }
1768 
1769     g_strfreev(lines);
1770 }
1771 
1772 static void dump_init(DumpState *s, int fd, bool has_format,
1773                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1774                       int64_t begin, int64_t length, bool kdump_raw,
1775                       Error **errp)
1776 {
1777     ERRP_GUARD();
1778     VMCoreInfoState *vmci = vmcoreinfo_find();
1779     CPUState *cpu;
1780     int nr_cpus;
1781     int ret;
1782 
1783     s->has_format = has_format;
1784     s->format = format;
1785     s->written_size = 0;
1786     s->kdump_raw = kdump_raw;
1787 
1788     /* kdump-compressed is conflict with paging and filter */
1789     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1790         assert(!paging && !has_filter);
1791     }
1792 
1793     if (runstate_is_running()) {
1794         vm_stop(RUN_STATE_SAVE_VM);
1795         s->resume = true;
1796     } else {
1797         s->resume = false;
1798     }
1799 
1800     /* If we use KVM, we should synchronize the registers before we get dump
1801      * info or physmap info.
1802      */
1803     cpu_synchronize_all_states();
1804     nr_cpus = 0;
1805     CPU_FOREACH(cpu) {
1806         nr_cpus++;
1807     }
1808 
1809     s->fd = fd;
1810     if (has_filter && !length) {
1811         error_setg(errp, "parameter 'length' expects a non-zero size");
1812         goto cleanup;
1813     }
1814     s->filter_area_begin = begin;
1815     s->filter_area_length = length;
1816 
1817     /* First index is 0, it's the special null name */
1818     s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1819     /*
1820      * Allocate the null name, due to the clearing option set to true
1821      * it will be 0.
1822      */
1823     g_array_set_size(s->string_table_buf, 1);
1824 
1825     memory_mapping_list_init(&s->list);
1826 
1827     guest_phys_blocks_init(&s->guest_phys_blocks);
1828     guest_phys_blocks_append(&s->guest_phys_blocks);
1829     s->total_size = dump_calculate_size(s);
1830 #ifdef DEBUG_DUMP_GUEST_MEMORY
1831     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1832 #endif
1833 
1834     /* it does not make sense to dump non-existent memory */
1835     if (!s->total_size) {
1836         error_setg(errp, "dump: no guest memory to dump");
1837         goto cleanup;
1838     }
1839 
1840     /* get dump info: endian, class and architecture.
1841      * If the target architecture is not supported, cpu_get_dump_info() will
1842      * return -1.
1843      */
1844     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1845     if (ret < 0) {
1846         error_setg(errp,
1847                    "dumping guest memory is not supported on this target");
1848         goto cleanup;
1849     }
1850 
1851     if (!s->dump_info.page_size) {
1852         s->dump_info.page_size = qemu_target_page_size();
1853     }
1854 
1855     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1856                                      s->dump_info.d_machine, nr_cpus);
1857     assert(s->note_size >= 0);
1858 
1859     /*
1860      * The goal of this block is to (a) update the previously guessed
1861      * phys_base, (b) copy the guest note out of the guest.
1862      * Failure to do so is not fatal for dumping.
1863      */
1864     if (vmci) {
1865         uint64_t addr, note_head_size, name_size, desc_size;
1866         uint32_t size;
1867         uint16_t guest_format;
1868 
1869         note_head_size = dump_is_64bit(s) ?
1870             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1871 
1872         guest_format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1873         size = le32_to_cpu(vmci->vmcoreinfo.size);
1874         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1875         if (!vmci->has_vmcoreinfo) {
1876             warn_report("guest note is not present");
1877         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1878             warn_report("guest note size is invalid: %" PRIu32, size);
1879         } else if (guest_format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1880             warn_report("guest note format is unsupported: %" PRIu16, guest_format);
1881         } else {
1882             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1883             cpu_physical_memory_read(addr, s->guest_note, size);
1884 
1885             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1886             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1887                                                desc_size);
1888             if (name_size > MAX_GUEST_NOTE_SIZE ||
1889                 desc_size > MAX_GUEST_NOTE_SIZE ||
1890                 s->guest_note_size > size) {
1891                 warn_report("Invalid guest note header");
1892                 g_free(s->guest_note);
1893                 s->guest_note = NULL;
1894             } else {
1895                 vmcoreinfo_update_phys_base(s);
1896                 s->note_size += s->guest_note_size;
1897             }
1898         }
1899     }
1900 
1901     /* get memory mapping */
1902     if (paging) {
1903         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1904         if (*errp) {
1905             goto cleanup;
1906         }
1907     } else {
1908         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1909     }
1910 
1911     s->nr_cpus = nr_cpus;
1912 
1913     get_max_mapnr(s);
1914 
1915     uint64_t tmp;
1916     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1917                        s->dump_info.page_size);
1918     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1919 
1920     /* init for kdump-compressed format */
1921     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1922         switch (format) {
1923         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1924             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1925             break;
1926 
1927         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1928 #ifdef CONFIG_LZO
1929             if (lzo_init() != LZO_E_OK) {
1930                 error_setg(errp, "failed to initialize the LZO library");
1931                 goto cleanup;
1932             }
1933 #endif
1934             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1935             break;
1936 
1937         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1938             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1939             break;
1940 
1941         default:
1942             s->flag_compress = 0;
1943         }
1944 
1945         return;
1946     }
1947 
1948     if (dump_has_filter(s)) {
1949         memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1950     }
1951 
1952     /*
1953      * The first section header is always a special one in which most
1954      * fields are 0. The section header string table is also always
1955      * set.
1956      */
1957     s->shdr_num = 2;
1958 
1959     /*
1960      * Adds the number of architecture sections to shdr_num and sets
1961      * elf_section_data_size so we know the offsets and sizes of all
1962      * parts.
1963      */
1964     if (s->dump_info.arch_sections_add_fn) {
1965         s->dump_info.arch_sections_add_fn(s);
1966     }
1967 
1968     /*
1969      * calculate shdr_num so we know the offsets and sizes of all
1970      * parts.
1971      * Calculate phdr_num
1972      *
1973      * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1974      * sh_info is 32 bit. There's special handling once we go over
1975      * UINT16_MAX - 1 but that is handled in the ehdr and section
1976      * code.
1977      */
1978     s->phdr_num = 1; /* Reserve PT_NOTE */
1979     if (s->list.num <= UINT32_MAX - 1) {
1980         s->phdr_num += s->list.num;
1981     } else {
1982         s->phdr_num = UINT32_MAX;
1983     }
1984 
1985     /*
1986      * Now that the number of section and program headers is known we
1987      * can calculate the offsets of the headers and data.
1988      */
1989     if (dump_is_64bit(s)) {
1990         s->shdr_offset = sizeof(Elf64_Ehdr);
1991         s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
1992         s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
1993     } else {
1994         s->shdr_offset = sizeof(Elf32_Ehdr);
1995         s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
1996         s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
1997     }
1998     s->memory_offset = s->note_offset + s->note_size;
1999     s->section_offset = s->memory_offset + s->total_size;
2000 
2001     return;
2002 
2003 cleanup:
2004     dump_cleanup(s);
2005 }
2006 
2007 /* this operation might be time consuming. */
2008 static void dump_process(DumpState *s, Error **errp)
2009 {
2010     ERRP_GUARD();
2011     DumpQueryResult *result = NULL;
2012 
2013     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2014         create_win_dump(s, errp);
2015     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2016         create_kdump_vmcore(s, errp);
2017     } else {
2018         create_vmcore(s, errp);
2019     }
2020 
2021     /* make sure status is written after written_size updates */
2022     smp_wmb();
2023     qatomic_set(&s->status,
2024                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2025 
2026     /* send DUMP_COMPLETED message (unconditionally) */
2027     result = qmp_query_dump(NULL);
2028     /* should never fail */
2029     assert(result);
2030     qapi_event_send_dump_completed(result,
2031                                    *errp ? error_get_pretty(*errp) : NULL);
2032     qapi_free_DumpQueryResult(result);
2033 
2034     dump_cleanup(s);
2035 }
2036 
2037 static void *dump_thread(void *data)
2038 {
2039     DumpState *s = (DumpState *)data;
2040     dump_process(s, NULL);
2041     return NULL;
2042 }
2043 
2044 DumpQueryResult *qmp_query_dump(Error **errp)
2045 {
2046     DumpQueryResult *result = g_new(DumpQueryResult, 1);
2047     DumpState *state = &dump_state_global;
2048     result->status = qatomic_read(&state->status);
2049     /* make sure we are reading status and written_size in order */
2050     smp_rmb();
2051     result->completed = state->written_size;
2052     result->total = state->total_size;
2053     return result;
2054 }
2055 
2056 void qmp_dump_guest_memory(bool paging, const char *protocol,
2057                            bool has_detach, bool detach,
2058                            bool has_begin, int64_t begin,
2059                            bool has_length, int64_t length,
2060                            bool has_format, DumpGuestMemoryFormat format,
2061                            Error **errp)
2062 {
2063     ERRP_GUARD();
2064     const char *p;
2065     int fd;
2066     DumpState *s;
2067     bool detach_p = false;
2068     bool kdump_raw = false;
2069 
2070     if (runstate_check(RUN_STATE_INMIGRATE)) {
2071         error_setg(errp, "Dump not allowed during incoming migration.");
2072         return;
2073     }
2074 
2075     /* if there is a dump in background, we should wait until the dump
2076      * finished */
2077     if (qemu_system_dump_in_progress()) {
2078         error_setg(errp, "There is a dump in process, please wait.");
2079         return;
2080     }
2081 
2082     /*
2083      * externally, we represent kdump-raw-* as separate formats, but internally
2084      * they are handled the same, except for the "raw" flag
2085      */
2086     if (has_format) {
2087         switch (format) {
2088         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB:
2089             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2090             kdump_raw = true;
2091             break;
2092         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO:
2093             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2094             kdump_raw = true;
2095             break;
2096         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY:
2097             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2098             kdump_raw = true;
2099             break;
2100         default:
2101             break;
2102         }
2103     }
2104 
2105     /*
2106      * kdump-compressed format need the whole memory dumped, so paging or
2107      * filter is not supported here.
2108      */
2109     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2110         (paging || has_begin || has_length)) {
2111         error_setg(errp, "kdump-compressed format doesn't support paging or "
2112                          "filter");
2113         return;
2114     }
2115     if (has_begin && !has_length) {
2116         error_setg(errp, QERR_MISSING_PARAMETER, "length");
2117         return;
2118     }
2119     if (!has_begin && has_length) {
2120         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2121         return;
2122     }
2123     if (has_detach) {
2124         detach_p = detach;
2125     }
2126 
2127     /* check whether lzo/snappy is supported */
2128 #ifndef CONFIG_LZO
2129     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2130         error_setg(errp, "kdump-lzo is not available now");
2131         return;
2132     }
2133 #endif
2134 
2135 #ifndef CONFIG_SNAPPY
2136     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2137         error_setg(errp, "kdump-snappy is not available now");
2138         return;
2139     }
2140 #endif
2141 
2142     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
2143         && !win_dump_available(errp)) {
2144         return;
2145     }
2146 
2147     if (strstart(protocol, "fd:", &p)) {
2148         fd = monitor_get_fd(monitor_cur(), p, errp);
2149         if (fd == -1) {
2150             return;
2151         }
2152     } else if  (strstart(protocol, "file:", &p)) {
2153         fd = qemu_create(p, O_WRONLY | O_TRUNC | O_BINARY, S_IRUSR, errp);
2154         if (fd < 0) {
2155             return;
2156         }
2157     } else {
2158         error_setg(errp,
2159                    "parameter 'protocol' must start with 'file:' or 'fd:'");
2160         return;
2161     }
2162     if (kdump_raw && lseek(fd, 0, SEEK_CUR) == (off_t) -1) {
2163         close(fd);
2164         error_setg(errp, "kdump-raw formats require a seekable file");
2165         return;
2166     }
2167 
2168     if (!dump_migration_blocker) {
2169         error_setg(&dump_migration_blocker,
2170                    "Live migration disabled: dump-guest-memory in progress");
2171     }
2172 
2173     /*
2174      * Allows even for -only-migratable, but forbid migration during the
2175      * process of dump guest memory.
2176      */
2177     if (migrate_add_blocker_internal(&dump_migration_blocker, errp)) {
2178         /* Remember to release the fd before passing it over to dump state */
2179         close(fd);
2180         return;
2181     }
2182 
2183     s = &dump_state_global;
2184     dump_state_prepare(s);
2185 
2186     dump_init(s, fd, has_format, format, paging, has_begin,
2187               begin, length, kdump_raw, errp);
2188     if (*errp) {
2189         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2190         return;
2191     }
2192 
2193     if (detach_p) {
2194         /* detached dump */
2195         s->detached = true;
2196         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2197                            s, QEMU_THREAD_DETACHED);
2198     } else {
2199         /* sync dump */
2200         dump_process(s, errp);
2201     }
2202 }
2203 
2204 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2205 {
2206     DumpGuestMemoryCapability *cap =
2207                                   g_new0(DumpGuestMemoryCapability, 1);
2208     DumpGuestMemoryFormatList **tail = &cap->formats;
2209 
2210     /* elf is always available */
2211     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2212 
2213     /* kdump-zlib is always available */
2214     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2215     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB);
2216 
2217     /* add new item if kdump-lzo is available */
2218 #ifdef CONFIG_LZO
2219     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2220     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO);
2221 #endif
2222 
2223     /* add new item if kdump-snappy is available */
2224 #ifdef CONFIG_SNAPPY
2225     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2226     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY);
2227 #endif
2228 
2229     if (win_dump_available(NULL)) {
2230         QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2231     }
2232 
2233     return cap;
2234 }
2235